Pat Simmons starts from a practical coordination problem: multiple AI agents, repositories and chat sessions become difficult to manage when every tool uses a different interface. He sketches a desktop environment with agent categories on the left and independent terminal tabs for each task.
Pat Simmons uses an AI coding model to turn the sketch into a Mac application through several feedback cycles. The resulting harness can create agent workspaces, open separate terminal sessions, connect subscription-based coding agents and route other models through an API provider without merging their working directories.
Pat Simmons then adds instruction and memory files so each agent has a defined role and can retain useful context between sessions. The demonstration shows why an agent environment needs explicit workspace boundaries, model configuration, credential handling and durable state in addition to a convenient interface.
Watch on YouTube
